Automatic blood vessel extraction for CTA images

When vascular structure information is segmented from three-dimensional data of Computed Tomography Angiography(CTA), it usually involves considerable amount of human intervention. To improve the vessel extraction efficiency, a fully automatic extraction method was proposed to segment blood vessels from three-dimensional data sets. Firstly, a multi-scale enhancement filter was developed to enhance the tubular-like structures, by which the non-tubular structures and noise were effectively removed. Then, a gradient image was combined with Sigmoid function to produce the speed image, and the Geodesic Active Contour(GAC) level set was utilized to approximate the real three dimensional vascular outline. Thereafter, the obtained vasculatures were processed by Laplacian smoothing function and a smoothed vascular surface was obtained. The proposed method was validated on both chest and neck CTA data. Experimental results show that blood vessels can be segmented accurately and automatically without human intervention. According to the phantom experiments, the average errors estimated for centerline and diameter of extracted vessels are 0.26 mm and 0.16 mm respectively. As there is no human interaction involved in the segmentation, the developed method can be utilized for the computer-assisted diagnosis of vascular related diseases in clinical practices.